Effective infrared spectral identification method via improved local binary patterns and spectral description representation

Abstract

Infrared spectral identification task which is an essential branch of infrared spectroscopy, has been extensively studied. In this article, we proposed an efficient description representation model with the local binary pattern (LBP) strategy, and applied to infrared spectral searching and identification. The idea is motivated by the fact that spectral curve can be considered as a composition of micro-patterns such as peaks, troughs, valley and flat lines which are well described by LBP. The whole spectral lines are split into several small sections, and then extracted the spectral LBP features. The method is capable of extracting both the local and holistic spectral features and robust to random noise and baseline interferences. Experimental results on the public infrared spectral dataset demonstrate that the proposed spectral identification method can significantly raise performance when noise and baseline interferences are present.